基于贝叶斯网络的流域实时防洪调度风险分析研究

It is of academic significance and important value to estimate and analyze the risk involved in real-time flood control operations of a river basin. Traditional risk analysis methods try to obtain high-dimension and multi-parameter joint probability density distribution functions of random variables, which is complicate and time-consuming. Furthermore, when being used in assessing flood risks, these methods attempt to formulate different joint probability density functions and recalibrate parameters for different flood events, limiting the model applicability and model timeliness. We will try to apply the Bayesian networks to analyze the synthesized risk of real-time flood control operation of a river basin. We will establish risk analysis models for real-time flood control operation of a river basin using Bayesian networks, including the risk prediction, risk assessment , risky decision-making models, uncertainty factors selection model and flood risk simulation model. We will propose the learning algorithms for Bayesian network parameters considering different data missing situations, such as extreme flood risk data missing and incomplete flood risk data. Besides, we will formulate online learning algorithms for re-adjusting dynamic Bayesian network parameters with real-time observations. The risk analysis method uses the Bayesian networks to disaggregate a multi-factor, multi-parameter and nonlinear joint probability density distribution function into multiple independent or conditionally correlated prior probability and conditional probability density distribution functions. It not only reduces the dimensionality and complexity of the flood risk analysis problem, but also improves the model applicability and meets the timeliness requirement of the risk analysis for joint operations of large-scale and complicated flood control systems in a river basin.

对流域实时防洪调度各环节中的不确定性及其带来的风险进行分析评估，具有重要的学术意义和实用价值。传统的风险分析方法大多需要求解高维、多参数的联合概率分布函数，其求解复杂、耗时，并且对于不同风险事件需要重新推求联合概率密度函数和率定参数，实时性和适应性较差，应用受限。本研究将贝叶斯网络理论用于流域实时防洪调度风险分析中，建立基于贝叶斯网络的流域实时防洪调度风险分析模型，贝叶斯网络风险预测、诊断和决策推理模型，风险因子筛选和风险情景仿真模型等；提出防洪风险资料极度缺乏、部分不完备和完备等情形下的贝叶斯网络参数学习方法，以及耦合实时校正成果的动态贝叶斯网络在线参数学习方法等。将多因子交织组合、多参数、非线性风险联合分布求解问题，分解为多个独立或条件关联的风险因子的先验概率和条件概率求解，降低了流域防洪风险求解的维数和复杂度，实时性和适应性强，满足复杂背景下大规模防洪系统联合调度风险分析的要求。

对流域实时防洪调度各环节中的不确定性及其带来的风险进行分析评估，具有重要的学术意义和实用价值。本课题提出了流域实时防洪调度过程中主要风险源的识别和风险因子筛选方法，构建了流域实时防洪调度过程中主要风险因子集，辨识了流域实时防洪调度风险因子之间的主次关系，进行了主要风险因子的时空演变规律分析；构建了流域实时防洪调度风险分析贝叶斯网络模型，并进行了贝叶斯网络的参数学习，既可以分析各种风险之间的相互影响关系，也可以为流域实时防洪调度的风险管理提供更加通用的管理工具和平台；基于贝叶斯网络等进行了流域实时防洪调度风险分析与决策方法研究；提出了在给定风险因子发生概率的情况下整个流域防洪系统或部分防洪系统风险的预测方法，提出了在给定防洪系统风险值的情况下，引发该风险的主要致险因子的诊断方法，建立了流域实时防洪调度的风险决策方法。